This paper presents a 7 year climatology describing medium-scale gravity waves observed in the menopause region covering the years from 1995 to 2001. The data comes from the OI and OH airglow observations of the three-field photometer employed at the University of Adelaide's Buckland Park, Australia (34.5°S, 138.5°E). About 1300 gravity wave events (AGW) are identified during the years 1995–2001. These AGW events usually persist for between 40 min and 4 hours. The magnitudes range from 1% to 14% of the background intensities and peak at 2% for OI observations and at 3% for OH observations. The observed periods range from 10 to 30 min, and the horizontal phase speeds range from 20 to 250 m s−1, with dominant wave scales of 17 min, 70 m s−1 for OI observations and 20 min, 40 m s−1 for OH observations. The intrinsic parameters are obtained by using medium-frequency (MF) wind data observed at the same place. The occurrence frequency of AGW events peaks at 13 min, 40 m s−1 for both OI and OH observations. The occurrence rate of gravity waves has a major peak in summer and a minor peak in winter. There is an obvious dominating southeastward direction for gravity waves, with azimuths of 160° in summer and 130° in winter. Studies for gravity waves observed in various locations show a similar tendency of propagating meridionally toward the summer pole. This implies that the tendency of propagating toward the summer pole may be a global trend for medium-scale gravity waves observed in the mesopause region. During summer, gravity waves propagate against winds measured by MF radar in their dominating direction. Using the ray tracing method, we found that the seasonal variation of winds limits the access of gravity waves to the observation height through reflection and critical coupling, which is one of the causes leading to the seasonal behavior of gravity waves observed over Adelaide.
 In order to understand the propagating mechanism and features of medium-scale gravity waves in upper atmosphere, many authors at different locations have carried out long-term continuous observations of gravity waves in the mesosphere and mesopause region. Giers et al.  presented the occurrence characteristics of gravity waves at 51°N through the 3 year observation of OH airglow. The observation of Giers et al.  shows mesospheric gravity waves in the range of 30–190 min for wave periods and 10–70 m s−1 for phase speeds, with the preferred direction toward the northwest. Wiens et al.  analyzed the 11 day data from the observation of O2 nightglow and found that the relation between frequency and the horizontal wavenumber was scattered rather than clearly linear and that absorption of westward traveling waves at speeds <30 m s−1 was evident. Walterscheid et al.  analyzed 9 months of observations of airglow emissions in Adelaide, Australia (34.5°S, 138.5°E). He found that the occurrence rate of gravity waves has a major peak in summer, which mainly propagates southward. Walterscheid et al.  interpreted the waves as ducted waves caused by the combined effects of Doppler and thermal ducting. Tang et al.  analyzed the 32 nights' OH observation data observed at New Mexico (35°N, 107°W) and reported that the seasonal direction of momentum flux is predominantly southwestward in winter and northwestward in summer. Both Tang et al.'s  and Walterscheid et al.'s  studies reveal a tendency for medium-scale gravity waves to propagate toward the summer pole. Manson et al.  reported the MF radar observation of mesospheric gravity waves in Saskatoon (49°–52°N) and Tromsø (69°N). The analysis shows that Saskatoon is dominated by meridionally propagating waves, while at Tromsø, zonal propagation dominates. The dominating direction of gravity waves differs from Tang et al.'s  and Walterscheid et al.'s  conclusions. Use of different instrumental measurements may be the cause of their different results as different instrumental measurements detect different parts of gravity wave spectrum.
 In this paper we report a 7 year statistical study of medium-scale gravity waves observed by a three-field photometer located at the University of Adelaide's Buckland Park (34.5°S, 138.5°E). We conduct a cross-spectral analysis to get the gravity wave propagation features from both OH (730 nm) and OI (557.7 nm) nightglow emissions. Time-averaged wind data were obtained using the MF radar, which also locates at Buckland Park. We used wind data to obtain intrinsic parameters of gravity waves. The characteristics of the gravity waves differ between the studies of Walterscheid et al.  and I. M. Reid and J. M. Woithe (unpublished manuscript, 2003). With the wind profile obtained from the HWM93 wind model [Hedin et al., 1991], seasonal behaviors are discussed. The study reveals some new features of medium-scale gravity waves in the mesopause region over Adelaide.
2. Data Processing
2.1. Instrumentation and Database
 Since 1995, a three-field photometer has been employed at the University of Adelaide's Buckland Park field to observe the airglow emissions (Reid and Woithe, unpublished manuscript, 2003). Two kinds of filters were installed: the 730 nm filters were used to observe the OH band emission, and the 557.7nm filters were optimized to observed the OI line emission at this wavelength. There are three observation positions in the photometer. Observation is conducted through the 557.7 nm filters on the half minute and through 730 nm filters on the minute. The center of the OH emission is between the heights of 85 and 87 km, with a full width half maximum of about 10 km, and the OI emission peaks at 97 km, with a half width of about 7 km (Reid and Woithe, unpublished manuscript, 2003). At the centre emission heights the field separation was 12 km for OH emission and 13 km for OI emission. Almost continuous data has been collected since 1995. We have the data for the years 1995–2001.
 There is no automatic way of detecting cloudy days. As Reid and Woithe (unpublished manuscript, 2003) reported, the data observed on cloudy days are characterized either by a high degree of correlation at zero lag for all the three time series or by many discontinuities from record to record (caused by varying transparency of passing clouds). On the basis of these characteristics, we excluded the data observed on cloudy days.
 An example of the intensity time series of airglow emission is presented in Figure 1, where the observation from three fields is shown. It is seen from Figure 1 that there are obvious medium-scale fluctuation structures in the original time series. As the relative locations of the three observation points are known, we are able to obtained observed wave parameters using spectral analysis.
2.2. Obtaining Wave Frequency and Phase Difference
 We adopt a three-channel maximum entropy method to get the wave frequency and phase differences between airglow intensity time series:
where I1, I2, and I3 are intensity time series observed at the same time at three fields. The maximum entropy method is represented by MEM. The method was introduced in detail by Strand . The variable f is wave frequency, and ϕ21 and ϕ31 are phase differences between time series observed at the three fields.
 The problem of phase ambiguity arises when calculating phase difference. The distance between three observation points is 12–13 km. For waves with horizontal wavelengths shorter than twice this distance the phase difference between the wave structures obtained from the maximum entropy method has the ambiguity of 2nπ (n is a unknown integer). This does not influence the determination of wave frequency.
 In this case, we have
where ϕ21 and ϕ31 are “real” phase differences between wave structures, ϕ21 and ϕ31 are phase differences obtained by the maximum entropy method, and m and n are unknown cycle numbers.
 We carried out the correlation analysis to determine the cycle numbers m and n. Earlier scientists reported the correlation analysis [Briggs, 1968] and similar fading method [Li, 1982] to determine the velocity of moving patterns. These methods analyze the phase difference directly in the time domain and thus can avoid phase ambiguity, although the phase difference obtained by these methods is the mean phase difference between the whole wave structure, and its accuracy is not satisfying. On the other hand, the value of ϕ21 and ϕ31 obtained from the maximum entropy method is the accurate phase difference except for the ambiguity of 2nπ and 2mπ. So, we calculated the cross-correlation coefficients between the observed time series. The lag corresponding to the biggest cross-correlation coefficient represents the time delay between wave structures. Thus we get the rough value of ϕ21 and ϕ31; using equation (2), we can estimate the cycle number m and n. Then, we also use equation (2) to get the “real” phase difference ϕ21 and ϕ31.
 An example is given in Figure 2, which shows the phase difference obtained from the maximum entropy method (Figure 2a) and the correlation analysis (Figure 2b). The variation of the observation height may cause an error while the phase difference is evaluated. The error bar is drawn in Figure 2 under the assumption that the observation height changes by 10 km above and under the center height. The phase difference ϕ21 is similar to ϕ21, while a 2π discrepancy is found between ϕ31 and ϕ31. Thus we get m = 0 and n = −1. We get the temporal variation of the phase difference through equation (2). The wave frequency and phase difference will be used in section 2.3 to determine other wave parameters.
 When more than one wave is observed at the same time series, we can only use the above mentioned technique to estimate the phase difference of the “main” wave with the highest energy in frequency spectra. In this case, the waves influence each other and give rise to phase chaos in the time series. The phase difference obtained by the correlation analysis is the mean phase difference for these waves. Thus we cannot determine m and n for other minor simultaneous waves. Phase chaos also influences the accurate determination of the “main” wave.
2.3. Evaluation of Wave Parameters
 A two-dimensional Cartesian coordinate system is set at the observation height, in which the wave parameter satisfies
where kx and ky are unknown horizontal wave numbers. Terms (x21, y21) and (x31, y31) are the relative positions of three observation points. Equation (3) yields the solution of kx and ky:
We consequently get the horizontal phase speed vph and wave propagation azimuth σ:
2.4. Data Processing
 The data processing consisted of three steps, following the use of the preprocessor as outlined by Reid and Woithe (unpublished manuscript, 2003). First, a sliding 40 min window was applied to the original time series at intervals of 2 min. This generated a sequence of shorter time series of duration 40 min. Background trends were removed from these 40 min time series using residues obtained from second-order polynomial fits. Second-order polynomials were chosen as they were sufficient to remove background trends in these short time series without introducing artificial perturbations as higher-order polynomials could do. The 40 min window length was chosen as we were only concerned with gravity waves whose period was 30 min or less, and a time series of 40 min is long enough to contain at least one cycle of such waves. Gravity waves with periods greater than 40 min are not included in this study as they are often mixed with background variations such as those produced by the Milky Way (reported by Reid and Woithe (unpublished manuscript, 2003) to have a temporal scale of 1 hour or more). Long-period trends are easy to remove from short time series such as those utilized here.
 Second, a cross-spectral analysis was carried out on each 40 min time series using the three-channel maximum entropy method. This generated wave frequency f, ϕ21, and ϕ31. The correlation analysis was applied to the time series in two successive windows to obtain the phase difference for wave envelopes. Time series in two windows were chosen to include several cycles of such waves. As this phase difference roughly equals ϕ21 and ϕ31, the cycle numbers m and n in equation (2) can be estimated. Then, we used ϕ21 and ϕ31 and m and n to get the precious value of phase difference ϕ21 and ϕ31 by equation (2).
 Finally, phase speed and wave propagation azimuth were obtained through equations (4)–(7). If there were wave-like structures in the time series, this analysis revealed the observed period, horizontal phase speed, and propagation direction of the structures in each time series. Since each time series was separated by 2 min, the temporal variation of wave parameters across a night's observation was revealed. If a steady wave field was passing, the wave parameters for components of the field would either remain steady for the duration of that field or vary in a well-defined way due to interference between different waves.
 To identify an AGW event, two conditions must have been satisfied: (1) the temporal variation of calculated wave periods and horizontal phase speeds should not change by more than 25%, and the change of wave azimuth should not exceed 10%; and (2) the duration of the gravity wave should be longer than 40 min. A considerable change of wave parameters sometimes occurs when the wave structure is influenced by temporal changes of wind shear, by the broken pieces of other gravity waves, or by the behavior of wave sources. Usually, the temporal variation of background winds causes the wave periods and phase speeds to change greatly. At the same time, wave azimuth will not be affected by the temporal variation of winds, although it is affected by other features such as movement of source or the broken pieces of other gravity waves. As the wave periods are limited to 40 min, only waves with time durations longer than 40 min can be considered to last at least one cycle in the observed time series. It is possible that “wave structures” lasting shorter than 40 min are, in fact, random changes of the observed time series. So, these conditions ensure the accurate calculation of wave parameters and the reliable identification of AGW activities.
 An example of the result is presented in Figure 3, which shows a gravity wave event detected between 0240 LT and 0340 LT in OH nightglow observations. The period was ∼30 min, the horizontal phase speed was 8 m s−1, and the direction of propagation was almost southward. The temporal change of wave period, phase speed, and propagation direction was not noticeable during passage of the wave, confirming the correct detection of an event.
 Error bars in Figure 3a were drawn based on the half width of wave frequency calculated by the maximum entropy method. It is shown that the error for wave period is about 5 min for the mean value, 8 min at most. The error may be caused by the temporal variation of background winds, which change the observed frequency of wave structures through Doppler shifting.
 Error bars in Figures 3b and 3c show the influence of variation in observation height. This kind of error first affects the phase difference (which has been shown in Figure 2), and then the error is transferred to phase speed and wave azimuth through equations (4)–(6). Also included in the error bars are the errors from the detection of wave frequency. It is shown that these kinds of errors are not noticeable and can be ignored.
 Other possible errors during the determination of wave parameters include the windowing technique and the polynomial fit used. Since the background variation has the timescale of 1 hour or more (Reid and Woithe, unpublished manuscript, 2003), the second-order polynomial fits for time series in a 40 min time window are enough to remove the linear trends. However, the nonlinear variation in time series cannot be fully removed by the windowing technique and the polynomial fits. This will influence the determination of wave parameters, although the influence can be reduced by the two conditions in identifying an AGW event, as described above.
3. Statistical Results
3.1. Gravity Wave Periods, Phase Speeds, and Wave Azimuth
Figures 4 and 5 present statistical results of atmospheric gravity waves observed between 1995 and 2001. Both the OH and OI observations are shown. Approximately 1300 gravity wave events were identified from OH and OI observations, with the observed period ranging from 10 to 30 min and the horizontal phase speed ranging from 20 to 250 m s−1 (Figures 4a and 4b). These AGW events usually persist for between 40 min and 4 hours. The dominant wave scale of the distribution is 17 min, 70 m s−1 for OI observations and 20 min, 40 m s−1 for OH observations. Horizontal wavelengths widely range from 10 km to over 200 km, dominating at 75 km for OI and at 50 km for OH.
 Mean wind data were used to obtain intrinsic parameters. The wind data were obtained using the MF radar that is also located at Buckland Park. The instrumentation and data processing was described by Reid et al.  and Reid and Woithe (unpublished manuscript, 2003). The data include meridional and zonal winds averaged from 1930 to 0530 LT. Figures 4c and 4d present intrinsic parameters for gravity waves where the corresponding wind data are available. In total, 881 AGW events are shown in Figure 4c, and 841 are shown in Figure 4d. The figures show little difference for intrinsic parameters evaluated from OH and OI observations. This implies that the Doppler shift of neutral winds causes the difference of observed parameters for OH and OI observations. Intrinsic periods range from 4 to 25 min, and intrinsic speeds are <150 m s−1. The occurrence rate peaks at 12 min, 40 m s−1 for both OI and OH observations. Gravity waves with an intrinsic phase speed >150 m s−1 are not found. The intrinsic periods are obviously shorter than the observed ones. Very few waves with an intrinsic period <5 min (approximate Brunt-Vaisala period at the observation height) are found. This may be caused by errors in our wave parameter evaluating method, which was presented in section 2. It may also be caused by errors arising in the calculation of intrinsic parameters when time-averaged winds are used.
 A similar gravity wave scale observed in OH nightglow was reported by Shiokawa et al. , who found gravity waves with phase speeds between 20 and 40 m s−1. Swenson et al.  reported a gravity wave with an intrinsic period of nearly 7 min and an intrinsic phase speed of 61 m s−1 at almost the same latitude. Walterscheid et al.  also analyzed the nightglow data with a duration of 9 months observed at Adelaide. He reported a typical wave scale of 50–80 m s−1 for measured phase speeds and 5–20 min for observed periods. The results shown in Figure 4 differ considerably from what was reported by Reid and Woithe (unpublished manuscript, 2003). Reid and Woithe (unpublished manuscript, 2003) also used the cross-spectral methods on the same database to calculate wave parameters. Their analysis yields gravity waves with periods widely ranging from several minutes up to 2 hours. Though there are abundant short-period waves in Reid and Woithe's (unpublished manuscript, 2003) results, nearly half of the waves are with periods longer than 30 min. The discrepancy of the two results is caused by the method we used. In Reid and Woithe's (unpublished manuscript, 2003) study the cross-spectral analysis was applied to the entire airglow intensity time series observed throughout the night. In this case, the short-period perturbations for one to several cycles would contribute little to the entire spectrum. In the present paper we conducted the cross-spectral analysis on 40-min-long time series. We concentrate on gravity waves with periods smaller than 30 min as the time window is set to be 40 min. There are one to several cycles included in the 40 min time series. While the method employed by Reid and Woithe (unpublished manuscript, 2003) was most sensitive to long-term perturbations, this present paper focused on shorter-term variations through the selection of a 40 min window. We obtain more accurate details about short-period waves, although obviously, information about waves with periods >40 min is lost.
Figure 6 presents the histograms of relative magnitudes. The magnitudes range from 1 to 14% of the background intensities and peak at 2% for OI observations and at 3% for OH observations. The average magnitudes coincide with those found by Gavrilov et al. , who reported the relative magnitudes of airglow variations averaged at 1.8–2.4% for emission rates at the same heights at (35°N). Shiokawa et al.  showed gravity waves with magnitudes beyond 10% in his case study. Waves with magnitudes >10% are also shown in Figure 6.
Figure 7 shows the year-to-year change of relative occurrence rates. The occurrence rates were normalized by total observation days. There is a peak of occurrence in 1998 and a minimum in 1997. We do not know the cause. In high years (2000 and 2001) the occurrence rates are just on average levels. This implies that the medium-scale gravity wave activities may have little to do with polar activities.
3.2. Seasonal Behaviors
 Previous studies conducted at Adelaide [Walterscheid et al., 1999; Hecht et al., 1997] have demonstrated a seasonal dependence of wave propagation directions. Figures 8 and 9 show the polar plots of seasonal change for the relative occurrence rate of gravity wave propagation azimuths. Occurrence rates were normalized with the number of observation days in each season to remove the bias caused by weather conditions. The gravity wave activity occurrence has a major peak in summer and a minor peak in winter. The occurrence rate is small in autumn and spring. The semiannual variation of gravity wave activities with maxima in winter and summer and minima in equinoxes has been addressed by many studies [Meek et al., 1985; Vincent and Fritts, 1987; Nakamura et al., 1996]. Nakamura et al.  showed that the semiannual variation might be caused by background wind variations.
 Polar plots in Figures 8 and 9 show a tendency for the gravity waves to propagate southward both in summer and winter. There is a weak eastward preference in summer. The eastward preference is relatively stronger in winter since the dominating directional azimuth is 160° in summer and 130° in winter. The dominating direction is similar to that reported by Reid and Woithe (unpublished manuscript, 2003). Walterscheid et al.  also found that there is a main peak of occurrence rate for poleward propagating gravity waves in summer. Recently, Nakamura et al.  analyzed the seasonal variation through an OH imager located at Shigaraki (35°N, 136°E) and found that the gravity waves are poleward propagating in summer and equatorward propagating in winter. Tang et al.  reported that the seasonal direction of momentum flux is predominantly southwestward in winter and northwestward in summer at New Mexico (35°N, 107°W). The results of Tang et al.  also show a weak westward preference both in summer and winter. All these results imply that the tendency of propagating meridionally toward the summer pole may be a global trend for medium-scale gravity waves observed in the mesopause region.
 Mean values of periods and phase speeds in four seasons are listed in Table 1. The values of observed phase speeds exceed 100 m s−1, except for the value in autumn for OH observations. Mean intrinsic phase speeds are around 50–60 m s−1. The intrinsic periods are also smaller than observed ones. This implies that the gravity wave has a preference of propagating against winds measured by MF radar. In summer, wave periods drop by about 3 min after Doppler correction, while in winter the drop is about 1 min. The result observed by OI observations is similar to that observed by OH.
Table 1. Mean Value of Periods and Horizontal Phase Speeds for Gravity Waves Observed by OI and OHa
To and Vpo are observed wave period and phase speed. Ti and Vpi are intrinsic period and phase speed after Doppler correcting using the MF radar mean wind data.
Vpo, m s−1
Vpi, m s−1
Vpo, m s−1
Vpi, m s−1
 In order to analyze the relation between gravity waves and background winds, histograms of the difference between the wave azimuths and wind-blowing azimuths are shown in Figures 10 and 11. Variables σwave and σwind are wave propagation azimuth and wind-blowing azimuth, respectively. It is found that the occurrence rate peaks in summer when the values σwave and σwind approach 180°, showing a strong tendency for gravity waves to propagate against background winds. This result is consistent with an early study of the wind filtering effect [Waldock and Jones, 1984]. The tendency of propagating against winds is not obvious in other seasons, although there is a minor peak of occurrence rate in winter. This may be caused by the seasonal variation of gravity wave sources.
 Source features also have important effects on the seasonal behavior of gravity wave activities. Walterscheid et al.  stated that the convections in the stratosphere might be one of the sources of gravity waves measured over Adelaide. Wu and Waters  found that medium-scale gravity waves observed in the middle atmosphere and mesosphere are strongly correlated with upper troposphere convection, surface topograph, and jet streams. All these possible sources contribute to the seasonal behavior of gravity wave activities since weather processes in the stratosphere are always season-dependent.
3.3. Effects of Mesospheric Zonal Winds
 Mesopheric winds have a profound influence on the propagation of medium-scale gravity waves. The mean zonal wind profiles obtained from the HWM93 wind model [Hedin et al., 1991] are presented in Figure 12. Meridional winds are not given for the reason that they are generally much weaker than zonal winds. MF radar winds are not used here. Since the wind data we have are averaged winds during night at the two observation heights, they cannot reflect the whole wind profile in which waves propagate. In winter the east blowing zonal winds dominate the troposphere and lower mesosphere. The wind reaches a peak (∼65 m s−1) at the height of 60 km. The winds grow weak above 60 km height. By contrast, the zonal winds in summer have a peak of 60 m s−1 westward at 60 km height. The wind profiles are similar to MF observations of zonal mean winds over Adelaide [Vincent and Fritts, 1987].
 Assuming that gravity waves are launched from the troposphere and propagate obliquely to the mesopause region, we adopt a ray tracing method and calculate ray paths of gravity waves propagating in the wind field showed in Figure 12. The ray tracing method was reported in detail by Ding et al. . Ray paths are given in Figure 13. Six subplots are presented, showing the propagation of gravity waves in summer and winter. The wave periods are set to be 8 min, 16 min, and 25 min, respectively, in both summer and winter plots. The phase speeds are 40 m s−1. In each subplot, seven rays are shown, with the initial wave azimuths ranging from 90° to 270°, with 30° intervals. The propagation of gravity waves with azimuths >270° or <90° is not presented for the reason that the dominating wave azimuth is not in these ranges.
 Mesospheric winds limit access of gravity waves to the observation height through reflection and critical coupling. In summer, as the zonal wind blows westward, waves with the azimuths of 90°, 120°, and 150° propagate against winds, and waves with the azimuths of 210°, 240°, and 270° propagate along winds. Waves with the azimuth of 180° are not affected by zonal winds. In summer, it is seen in Figure 13 that waves with a period of 8 min and azimuths >210° are critically coupled to the mean flow, and waves with azimuths <180° are reflected. Only waves propagating southward reach the observation height. For waves with periods of 16 min and 25 min, all the waves with azimuths <180° reach the observation height, while waves with azimuths >180° cannot. This is coincident with the statistical results in Figures 7–10, where the gravity waves in summer propagate against the dominating azimuth <180°. Wind speeds projected on the wave propagation directions are rather weak, which prevents the waves from being reflected.
 In winter, it is found in Figure 13 that the possible azimuths for waves that reach the observation height should be >120°. Since the dominating azimuth in winter is ∼130° (Figures 8 and 9), the waves observed in winter may have traveled a thousand kilometers before they reach the observation height.
 Saturation of gravity waves in the mesosphere was observed by the Upper Atmosphere Research Satellite [Wu and Waters, 1996]. The above mentioned ray tracing method cannot study gravity wave energy dissipation in the mesosphere region. For gravity waves propagating southeastward the horizontal range calculated by the ray-tracing method is longer in winter than in summer (Figure 13). This leads to more severe energy dissipation accumulated on ray paths in winter. This may be one of the reasons why the occurrence rate of gravity waves observed in the mesopause region is smaller in winter than in summer.
 The horizontal range shown in Figure 13 implies the distance between the gravity wave source and the observation height. It may not reflect the distance between the source and the photometer in Buckland Park. It has been suggested by previous studies [Walterscheid et al., 1999] that the gravity wave observed over Adelaide may be ducted waves trapped by a thermal duct and a Doppler duct. Shiokawa et al.  found through the observation of three all-sky airglow imagers that gravity waves in the mesopause region have a spatial extent of >1000 km. However, before gravity waves arrive at the mesopause region they are launched into the troposphere and propagate obliquely to the mesosphere region, passing through the mesospheric wind field. In this stage the ray-tracing method can be used to study gravity wave propagation features, as described in this section.
 This paper presents the 7 year statistical characteristics of medium-scale gravity waves observed by a three-field photometer in Adelaide (34.5°S, 138.5°E). The study reveals some new features of gravity waves at midlatitude in the Southern Hemisphere. The results are summarized as follows.
 1. There are abundant medium-scale gravity waves in the mesopause region over Adelaide. Approximately 1300 gravity wave events are identified from both OH observations and OI observations between 1995 and 2001. The relative magnitudes peak at 2% of emission intensities for OI observations and at 3% for OH observations. The year-to-year change of occurrence rate is generally not obvious.
 2. The observed periods range from 10 to 30 min, and the horizontal phase speeds range from 20 to 250 m s−1. Using the mean winds obtained from MF radar observations, we found that the intrinsic periods range from 4 to 25 min and that the horizontal phase speeds are <150 m s−1. The occurrence rate of gravity wave events has a peak at 13 min, 40 m s−1 both for OH and OI observations. The distributions of occurrence for OH observations are similar to those for OI observations.
 About 5% of gravity waves have intrinsic phase speeds <10 m s−1. The phase speeds of this range are near the background wind speed. In this case, it is possible that we are not observing a wave but rather structure that is propagating in the mean flow, such as a broken wave. It may also be caused by errors aroused when we used time-averaged winds in the calculation of intrinsic phase speeds.
 3. The occurrence rate of gravity waves has a major peak in summer and a minor peak in winter. The results show a tendency for the gravity waves to propagate southward both in summer and winter. There is a weak eastward preference in summer. The eastward preference is relatively stronger in winter. Studies for gravity waves observed in various locations show a similar tendency of propagating meridionally toward the summer pole [Walterscheid et al., 1999; Nakamura et al., 2001; Tang et al., 2002; Reid and Woithe, unpublished manuscript, 2003]. This implies that the tendency of propagating toward the summer pole may be a global trend for medium-scale gravity waves observed in the mesopause region.
 It is possible that the determination of propagation direction is influenced by phase ambiguity, as described in section 2.2. This happens when gravity waves are detected with a horizontal wavelength smaller than twice the distance between observation points (namely, 24 km for OH and 26 km for OI). It is shown in Figure 5 that the majority of the waves have a horizontal wavelength >24 km. Hence the influence of phase ambiguity on the statistical results of wave direction is negligible.
 During summer, there is a strong preference for gravity waves to propagate against winds measured by MF radar, while in winter, their preference is to propagate with winds. By the application of the ray-tracing method, we found that the seasonal variation of winds is one of the causes for the seasonal behavior of gravity waves through limiting access of gravity waves to the observation height by reflection and critical coupling. Source features are also discussed as another important cause of seasonal behavior for gravity wave activities.
 This work was supported by the Natural Science Foundation of China (40304011, 40134020, and 40374054) and the National Important Basic Research Project (G2000078407). We also acknowledge the Australian Research Council for supporting this work under grant number 20006300.